RoboSynChallenge evaluation weights
Inference checkpoints and runtime resources for all ten RoboSynChallenge tasks.
Code and setup: GitHub. Exact task-to-checkpoint paths: manifest.json. Required files and SHA256 hashes: artifacts.json.
After cloning the code and following its setup instructions, run in the code directory:
python download_assets.py
python verify_package.py --weights
python run_task.py water_pouring 0 --smoke
Replace the task name and GPU index as needed. Keep the original directory structure. The repository includes params, normalization statistics, checkpoint metadata, hybrid auxiliary weights, the tokenizer and simulation resources. No train_state, training data or temporary evaluation outputs are required.